Testing of Atmospheric Turbulence Effects on the ...€¦ · Drag coefficient Zero lift drag ... Over flat uniform terrain it typically extends to a height of 500-1000 ... and yaw

Testing of Atmospheric Turbulence Effects on

the Performance of Micro Air Vehicles

David Lundström and Petter Krus

Linköping University Post Print

N.B.: When citing this work, cite the original article.

Original Publication:

David Lundström and Petter Krus, Testing of Atmospheric Turbulence Effects on the

Performance of Micro Air Vehicles, 2012, INTERNATIONAL JOURNAL OF MICRO AIR

VEHICLES, (4), 2, 133-149.

http://dx.doi.org/10.1260/1756-8293.4.2.133

Copyright: Multi-Science Publishing

http://www.multi-science.co.uk/

Postprint available at: Linköping University Electronic Press

http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-81837

http://dx.doi.org/10.1260/1756-8293.4.2.133

http://www.multi-science.co.uk/

http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-81837

1

Testing of Atmospheric Turbulence Effects

on the Performance of Micro Air Vehicles

David Lundström and Petter Krus

Linköping University, Department of Management and Engineering

58183 Linköping, Sweden

Email: [email protected] / [email protected]

ABSTRACT

Micro Air Vehicles (MAV) are generally operated at low altitudes and within the earth

boundary layer. This is a very dynamic environment with varying wind intensity and

turbulence levels far greater than those experienced by traditional manned aircraft

cruising at higher altitudes. Yet aerodynamic research on MAVs is often based on the

assumption of steady aerodynamics. Little effort has been made to experimentally

determine the validity of this assumption. In this paper, the effect of turbulence on the

performance of a MAV is studied using flight testing in different wind conditions.

Flight testing technique, data logging equipment and data reduction are explained.

Additionally, a low cost technique for propeller performance measurement is

presented. Results show that the flow around a MAV flying in windy conditions

qualifies as highly unsteady, although the impact on its performance is surprisingly

small for the kind of turbulence levels at which MAVs can be expected to operate.

Accelerometer data from the flights reveals that if steady aerodynamic theory is

assumed, increasing turbulence should have resulted in a measurable drag increase,

thus indicating that the tested MAV to some extent passively manages to benefit from

the turbulence.

NOMENCLATURE AR Aspect Ratio

c Wing chord

CD Drag coefficient

Zero lift drag coefficient

Lift induced drag coefficient

CL Lift coefficient

CP Propeller power coefficient

CT Propeller thrust coefficient

D Drag

d Propeller diameter

J Advance ratio

k Reduced frequency parameter

L Lift

n Propeller revolutions per second

Ω Propeller radians per second

P Power

R0.75 Propeller ¾ radius length

S Wing reference area

T Thrust

v Free stream velocity

vRe Propeller Reynolds number velocity

ρ Air density

ω Angular frequency

Efficiency

1. INTRODUCTION

The atmospheric boundary layer (ABL) is the region of air affected by the friction between the

surface of earth and the wind. It results in a gradient wind field within which the air flow is

complex and turbulent. The characteristics of the ABL depend on the underlying terrain and

atmospheric conditions. Over flat uniform terrain it typically extends to a height of 500-1000

meters above the ground [1]. MAVs normally operate in the lower region of this envelope, where

the turbulence intensity is very strong, and atmospheric turbulence must therefore always be

expected to impact MAV aerodynamics. The characteristics of the atmospheric boundary layer

have been well studied in the past in the field of meteorology and wind engineering [1-3]. More

recent studies have explored the characteristics of the ABL very close to the ground and with

direct application to MAVs [4].

The practical implications of a MAV flying in turbulence is, most obviously, that its attitude

Testing of Atmospheric Turbulence Effects on the Performance of Micro Air Vehicles

2

and flight path will be disturbed. Turbulence or wind gusts will induce unwanted pitch, roll and

yaw inputs and trigger dynamic responses such as phugoid, short period and Dutch roll oscillations

[5]. Most research regarding the effect of atmospheric turbulence on MAVs is targeted towards

understanding and minimizing these unwanted responses, with the goal of making MAVs into as

stable of sensor platforms as possible.

An area that has been paid less attention is the effect of atmospheric turbulence on aerodynamic

performance. There are several reasons why turbulence would be expected to influence both lift

and drag properties. The constant attitude changes and excitation of oscillatory modes need to be

either corrected by the control system or dampened by the aerodynamic stability inherent in the

design. This results in extra drag. Moreover, the rapid fluctuations in flow direction in the ABL

have the potential to profoundly affect the aerodynamics of the vehicle. Measurements by Watkins

et al [4], taken 2 meters above the ground in 4 m/s wind, illustrate the complex nature of the flow

in the ABL (Figure 1).

Figure 1. Pitch angle variation in 4 m/s wind measured by four probes with 150 mm

lateral separation and at an elevation of 2 m, from Watkins et al [4], with permission.

For such rapid flow fluctuations, traditional stationary aerodynamic theory is invalid. At higher

altitudes and at typical flying speeds the variations in pitch angle will be less pronounced but it is

still likely that the aerodynamics of MAVs flying in the lower regions of the ABL will qualify as

unsteady. Assumptions regarding lift and drag made during design may not be valid. As long ago

as 1922 Katzmayr [6] showed that an oscillating airflow considerably altered the aerodynamic

properties of airfoils. His results showed that the average drag as a function of average lift could

increase as well as decrease depending on the conditions. The latter is often referred to as the

Katzmayr effect. Even though this is established knowledge, aerodynamic research on MAVs is

often based on the assumption of stationary flow. For instance, traditional wind tunnels do not

replicate the turbulent environment representative of the real world atmosphere [7]. Yet it is the

common tool for experimental MAV research. Similarly, in CFD simulations the common

assumption is that the inflow is smooth and uniform.

With the above considerations in mind, it was decided to experimentally study to what extent

the turbulence within the lower regions of the ABL affects the performance on MAVs. A

straightforward approach with flight testing was selected. One of the University MAVs was

equipped with data logging equipment and flight-tested in several different ABL conditions with

the objective to determine its aerodynamic efficiency. The assumption was that increasing wind

and turbulence would result in a noticeable change in performance. In addition to evaluating the

relative effect of turbulence, the flight tests was also intended to provide absolute data for latter

evaluation of MAV performance prediction methods as part of the authors’ on-going research in

MAV design automation [8, 9].

Aerodynamic efficiency, i.e. lift to drag, in flight can be measured either by means of a power

off glide slope technique or with power on using a thrust-calibrated propulsion system. For

turbulent conditions, the power on method is the only feasible method. Performance flight testing,

using this technique applied to small electric powered UAVs, has previously been demonstrated

by Ostler et al [10]. They calibrated thrust as a function of velocity and motor input power.

David Lundström and Petter Krus 3

Considering that electric motor efficiency is coupled to motor temperature [9], it was decided for

these tests to use the more common method of calibrating propeller characteristics as a function of

rpm and velocity.

2. PROPULSION SYSTEM CHARACTERIZATION

Propeller characteristics are commonly described by the thrust and power coefficients, CT and CP,

(Equation 1-2) as a function of advance ratio, J, (Equation 3).

Ideally, CT and CP are only dependent on advance ratio but in reality they are also coupled to

Reynolds number and aeroelastic effects such as blade twisting. In order to fully define the CT and

CP coefficients the propeller's thrust and input power need to be mapped against the entire velocity

and rpm range over which it is intended to operate. This is traditionally carried out in wind tunnel

testing. Examples of wind tunnel data for typical model hobby propellers can be found at the

University of Illinois extensive propeller database [11]. Unfortunately, this database contained no

data for propellers small enough to suit the MAV selected for the tests. As a first step, the

propeller of the MAV, an APC 6x5.5, had to be tested. With no wind tunnel available an

alternative test method was developed. As a wind tunnel substitute a measurement rig was

mounted on the roof of a car. The in-house designed test rig is shown in Figure 2. The rig

measures thrust and torque using a set of carefully arranged strain gauges. The measurement range

is 0-30 N of thrust and 0-0.3 Nm of torque, each sampled at 15 bit resolution. Additionally the rig

measures rpm, free stream static and dynamic pressure, air temperature, flow inclination in pitch

and yaw using vanes, and, although not required, it also measures the motor current and input

voltage. The hardware design for torque and thrust measurement is described in more detail in Ref

[12] where it was used to study the characteristics of electric motors and motor controllers.

Figure 2. Car-mounted propeller test rig.

The data logging was performed using an Eagle Tree Systems USB Flight Data Recorder Pro.

Together with additional sensors for measuring analog signals and GPS, it offered a convenient

off-the-shelf solution for data acquisition. The rig was mounted on top of a sturdy tripod which in

turn was mounted on the roof of the car. To minimize the aerodynamic disturbance from the car

the tripod places the propeller as high as two meters above the car roof. Both the data logger and

(3)

(1)

(2)


4

the motor controller were controlled from a laptop placed on the passenger seat. The test

procedure consisted of starting the logging, accelerating the car up to a given speed at which the

cruise control was activated, and then executing an automatic test sequence that advanced the

motor’s electronic speed controller (ESC) in steps of 12.5% until 100% power was reached. An

example of a data log from such a test is shown in Figure 3.

Figure 3. Raw data from propeller car top testing.

During the testing, factors such as wind, vibrations due to road roughness etc induced noise to the

measurements. To minimize the effect of this noise, the length of each segment with constant

throttle was 13 s. Along with a log frequency of 40Hz, this provided enough data samples to easily

filter out the noise. To capture any effects of Reynolds number and potential blade twisting, the

test procedure was repeated for 6 different speeds in the range of 40-100 km/h, as well as a static

test. The actual testing was carried out on a public road in flat terrain. To avoid traffic and to

obtain as low and constant wind conditions as possible, all the testing was performed at night.

In order to simplify the data reduction, a script was written that automatically screened through

the data to sort out useful data sections. The script divided the data into segments of 120 data

points. Within each segment, the script examined the data to verify that each parameter stayed

within predefined min/max values and that no variation larger than the expected noise occurred. If

all requirements were met, the data for that segment was averaged and stored in a new dataset.

This ensured that for the extracted data, the car was not under acceleration, the throttle did not

change, and no sudden changes in flow inclination or pitot pressure occurred. In order to visualize

and further validate the results of the data screening, a Google Earth KML file was generated

where the GPS trail is coloured depending on the quality of the data. Figure 4 illustrates the KML

file from one test. The green colour represents segments of valid data.

Figure 4. GPS trail of propeller test.

At the time of the tests, the weather conditions were what one generally would observe as calm.

However, weather services reported a constant breeze of approximately 1.5 m/s in a direction of

approximately 45 degrees to the road. This was observed in the data as an inclination of the free

stream relative to the propeller shaft, as well as a difference between airspeed and GPS speed. At

higher speeds, the flow inclination was negligible but at the minimal tested speed (40km/h) the

data log showed inclinations of approx. 8 degrees with peaks up to 11 degrees. When processing

the data, considerations were made about what effect the flow inclination would have on the


validity of the pitot tube measurements as well as the thrust and power measurements. The pitot

tube range of insensitivity to inclination has not been calibrated, but the design is made following

the recommendations made by Gracey [13] and should have an error margin less than 1% for flow

inclinations below 10 degrees. No corrections were therefore made to the airspeed measurements.

The effect of flow inclination on thrust and power coefficients have been well studied and

documented. Correction methods have been thoroughly described by for instance DeYoung [14].

For inclinations of less than 10 degrees however, not much correction is needed. This is also

supported by wind tunnel testing conducted by Leasley et al [15]. Since the subsequent MAV

flight testing was to be carried out without any direct flow inclination sensor, it was decided that

the flow inclination encountered during the propeller testing was actually positive since it would

be fairly representative of the angle of attack vs. flight speed the MAV would experience in flight.

It was therefore decided not to make any corrections regarding the flow inclination. A correction

was however made for the drag created by the motor and motor mount. This was done by

measuring the drag of the motor without any propeller mounted and then subtracting this from the

data. The total thrust as a function of rpm and velocity is shown in Figure 5 together with a

Matlab-generated surface fit.

Figure 5. Logged propeller thrust plotted against rpm and velocity.

The measured data of thrust and torque were recalculated to thrust and power coefficients. It was

seen that within the range over which the propeller would operate in flight, the propeller

performed consistently with little Reynolds number effects. Only at the lowest tested speed, and in

static testing, were the thrust and power coefficients seen to have some degradation. The resulting

curves, plotted for each velocity, are shown in Figure 6. As a reference, the Reynolds number in

these tests spans from 23,000 to 56,000, based on the blade chord at 75% of blade radius and with

a velocity defined as √ ( )

. These points correspond to the lowest rpm of the

lowest speed and the maximum rpm of the highest speed, which both lies at approximately J=0.8.


6

Figure 6. Thrust and power coefficients of APC 6x5.5 from propeller testing.

Based on these results, a Matlab surface fit model was created that was used for drag

determination.

3. FLIGHT TESTING

The objective of the flight testing was to determine lift to drag characteristics of the MAV for

different wind conditions, using the experimental propeller data. This means that any observed

difference in aerodynamic properties due to variations in atmospheric turbulence cannot be

distinguished whether it is derived from changes in the vehicles aerodynamics or propellers

performance. It is nevertheless representative of any performance variation due to turbulence

which was the purpose of the study.

3.1 Equipment

The aircraft selected for the flight test is an in-house designed Micro Air Vehicle called PingWing

(Figure 7), originally developed for the 2007 US-European MAV competition [16]. The PingWing

has been proven to perform well in very high winds and has forgiving flight characteristics in

turbulence. The specifications of the PingWing MAV are summarized in Table 1.

Figure 7. PingWing MAV


Table 1. Specifications of PingWing MAV.

For this particular test the PingWing was equipped with an Eagletreesystems elogger V4 data

logger, complemented with additional Eagletreesystems sensors for measuring airspeed, altitude,

GPS, 3-Axis acceleration and air temperature. No autopilot was used and all piloting was done

under manual control. The logging frequency was set to 50 Hz.

3.2 Test Procedure and Data Analysis

The testing consisted of flying the MAV in a fixed pattern of as long straight lines as possible

followed by gentle 180 degree turns. After take-off the pattern was entered at full throttle, which

was then gradually decreased in small arbitrary steps until the airspeed no longer allowed the

altitude to be maintained. For each throttle level, the aircraft was flown for two laps in the pattern.

Once the lowest possible throttle to remain at altitude was reached, the throttle was increased but

with a nose high attitude in order to further reduce the flight speed.

The flights were taking place in a relatively flat landscape. To give some sort of description of

the terrain of the test sites, the Davenport classification scheme [17], as summarized in Table 2,

has been used. The Davenport classification was developed in the discipline of wind engineering

and has previously been used by Watkins et al [4] to classify the environment of MAVs.

Table 2. Davenport classification of effective terrain roughness.

Some of the sensor data were somewhat noisy and seemed to be affected by electric noise. This

included the throttle signal, rpm, and current measurements. As a first step, a 50-point moving

average smoothing was applied to the data. An automatic screening was then, similarly to the

propeller testing, made to sort out time frames where the data was considered valid for further

analysis. By valid data denotes that the MAV was to be in level flight without any significant

variation in altitude, throttle, or GPS heading. Instinctively, it may seem as if there should have

been a requirement concerning variation in airspeed so that any acceleration after a turn is filtered

out, but strong variations in airspeed in turbulent conditions made any such filtering impossible.

On the other hand, it was found that the reduction in airspeed during turns was very small, and in

combination with the data from the long straight passes that followed would barely influence the

results. An example of flight trajectory is shown in Figure 8, where the valid time frames are

marked in green.


8

Figure 8. Example of flight trajectory used in testing. The green color marks segments of

valid data.

For the data in the valid time frames, lift and drag coefficients according to Equations (4 - 5) were

calculated for each data sample. The data was then averaged for each second, creating a smaller

data set that could be illustrated and plotted more easily. A second data set was also created where

the raw data was averaged for each step in throttle used during flight test, from full throttle down

to minimum required throttle to stay flying. These averaged points were finally used to curve fit

the lift to drag equation according to Equation (6).

In Equation (4), Lift, (L) is always considered to be equal to weight (mg) and in Equation (5), drag

(D), is equal to thrust (T) according to Equation (1).

4. RESULTS

Flights were made on several occasions and with varying wind conditions. In order to illustrate the

differences between turbulent and calm conditions, this paper presents the results from the

extremes: perfectly calm and extraordinary turbulent. All flights were conducted during November

and in similar conditions in terms of pressure and temperature. In all tests, the sky was overcast

and there was no noticeable thermal activity. In all, the results from three flights are presented.

Flight1: This flight was conducted in pretty much as ideal conditions as they can ever be. The

wind was exceptionally calm. The ground based wind meter at the site reported 0 m/s although

flight data revealed a slight breeze of approximately 1.5 m/s at altitude. Temperature was 8ºC,

humidity 82% and air density 1.26 kg/m3. This kind of calm wind conditions and total lack of

thermal activity rarely occurs. The terrain of the test site should according to Table 2 be

defined as #4 “Roughly open”. The results from this flight should be considered as a reference

of the most forgiving ABL conditions possible.

(6)

(4)

(5)


Flight 2: This flight was conducted at a day of high winds and at the same location as flight 1

(Davenport #4). The on-site wind meter reported 9m/s with gusts of up to 12 m/s. This can be

regarded as an approximate upper wind speed limit for operation of MAVs. Temperature was

8ºC with a humidity level of 80% and air density of 1.23 kg/m3. The flight was conducted in

long straight passes aligned with the head- and tail-wind respectively. The altitude of the flight

was approximately 40 meters above the ground. The conditions of this flight should be

considered rough but as to be expected at these wind speeds.

Flight 3: Flight 3 was conducted on the same day as Flight 2 but at a different location that

generated more turbulence. This location should be described as a 7 on the Davenport scale.

The site was situated next to a forest and with the wind direction perpendicular to the trees

ridge line. The flying was conducted parallel to the ridge line on the leeward side in order to

attain maximum turbulence. An approximate wind speed as acquired from flight data was 7-8

m/s at the flight altitude of around 1-1.5 times the trees’ height. This generated a massive

amount of turbulence to a degree that should be considered unrealistic in normal flight.

Temperature was 8ºC, humidity 85% and air density 1.23 kg/m3.

For each of the above flights the data was treated as described in the previous paragraph. Figure 9

illustrates the results from each of these flights. The multi-colored point clouds represent the one

second averaged data. This illustrates the spread of the data for the different flights. The color is

used to distinguish at which throttle level the data points were gathered. The green data points

represent the total average for each throttle level. The black lines represent the curve fit of the data

points. Finally, the curve fits for each of the three flights are compared in the lower right graph.

Figure 9. Data log result from each test flight. The scatter points represent one second

averaged data and is colored based on throttle position. The larger green dots represent

the average for each throttle level. The black curves are the polynomial fit of CL vs CD.

Lastly, in the lower right figure, the polynomial fit of each flight is compared.

Since there were no major difference in temperature and pressure between the three flights, the


10

results are suitably illustrated by plotting the raw data of airspeed and propeller rpm (Figure 10).

Figure 10. Raw data of airspeed as a function of propeller rpm.

The results show that even though the turbulence in Flight 1 and Flight 2 is at the opposite ends on

the scale of practically occurring turbulence in MAV flight, the resulting lift to drag, averaged

over time, is close to identical. This is the reason why only the results from the extremes are

presented.

The results raised some questions about the validity of the data. It was very clear that the same

rpm generated virtually the same indicated flight speed but uncertainties were if the airspeed was

incorrectly measured for the turbulent flight, or if the energy required to turn the propeller at the

same speed would be different for the turbulent condition. To validate the pitot tube

measurements, the airspeed for each throttle level was compared to the GPS speed averaged for

both up and down wind flight. This was done for the entire velocity range in both Flight 1 and

Flight 2. The result indicated no difference in the airspeed measurements. For Flight 3, “chaotic”,

it was not possible to validate the pitot tube measurement; the result from Flight 3 therefore has

less credibility but is included as a reference that nothing drastic happens even at unrealistic

turbulence levels. The power to the motor was included in the data log and comparisons between

the three flights revealed no measurable difference in power to rpm.

To illustrate the magnitude of turbulence, the Z-axis accelerometer data for a segment of

straight and level flight from each of the three flights are given in Figure 11. As a reference of

accelerometer noise due to motor vibrations or possible electronic noise, the top left chart shows

accelerometer data of static testing with the MAV suspended horizontally on flexible wires and

with the motor running at full speed.


Figure 11. Z-axis accelerometer data of the three flights.

Considering the 50hz log frequency and the fact that the response time, or filter properties, of the

Eagltree accelerometer unit is unknown, these data can only be seen as an indication of the

fluctuations in angle of attack. Higher frequencies and larger peaks may exist but fail to be

recognized by the logging equipment. It is interesting to note that even for the percieved calm air

there is still some vertical movement. From the pilot perspective, not the slightest occurrence of

wind or turbulence was noted. The air felt completely “dead” and no roll or yaw dirturbance,

typical of turbulence, was seen in the MAV. In order to further illustrate the influence of

turbulence on the MAV, complete 3-axes accelerometer data from a longer sequence of Flight 2 is

given in Figure 12.

Figure 12. 40 seconds of 3-axis accelerometer data from Flight 2, at a velocity of 22 m/s.

The most noteworthy of these data is the magnitude of the Y-axis accelerations, illustrating the


12

significant oscillatory disturbances that occur around the yaw axis. This is typical of MAVs where

the short distance between center of gravity and vertical stabilizer provides low yaw damping.

From the pilots perspective this is seen as a constant excitation of yaw oscillations, similar to

Dutch roll, which never sees enough smooth air to dampen out.

4.1 Complementary Testing

The focus of the flight testing was to compare the aerodynamic properties between different

turbulence levels; however, the instrumentation in the MAV allowed some other interesting results

to be extracted. In the above presented result lift to drag data is compared only down to the flight

speed of minimum required power. Below this speed, the power required increases, but for the

turbulent flights it was impossible to acquire reliable data in this low speed region. The entire lift

to drag envelope was however tested for the calm flight where the MAV easily could be flown

undisturbed at high angles of attack. Out of curiosity, the MAV was also flown the test sequence

inverted. This would give some indication of the airfoil’s impact on performance. The results over

the complete operating range, including inverted flight, are shown in Figure 13. The figure

includes data points averaged for each throttle level, as well as a CLCD curve fit. For the purposes

of comparison, the negative CL curve is also mirrored to the positive side, illustrated by the faint

grey line.

Figure 13. Total lift to drag characteristics including inverted flight. The left figure displays

the entire operating range. The right figure displays the data in the range between full

power down to minimum power required.

Interestingly for full power, as well as typical cruise speed, there is barely any difference in

performance between inverted and upright flying. At lower speed, however, the inverted

performance was degraded with increasing drag and lower CL max.

In the current testing no correction of lift and drag has been made to account for the effect of

tilting trust vector as angle of attack increases. This is because it is very difficult to measure angle

of attack of a MAV in flight. Since MAVs, due to their compact geometry and low aspect ratio,

greatly affect the adjacent flow field, a flow inclination device to measure the free stream angle

cannot be installed. It would, however, be interesting to have some quantifiable measure of within

what angle of attack range the MAV was operating. The PingWing was equipped with a 3-axis

accelerometer. As long as the aircraft remains in straight and level flight without any significant

turbulence, the relationship between the x- and z-axis accelerometers should provide a decent

value of angle of attack. This was tested in calm wind and for positive angle of attack. The result,

averaged for each step in throttle, is shown in Figure 14.


Figure 14. Lift coefficient as a function of angle of attack as acquired from accelerometer

measurements.

Due to the previously mentioned uncertainties in the accelerometer data, the results should be

interpreted with caution. The result however indicates that the MAV has an upper angle of attack

limit of approximately 18-20 degrees. Some nonlinear behaviour appears to occur for lower angle

of attack, while seen over the entire envelope the relationship is close to linear.

Another interesting complication with MAVs, compared to larger aircraft, is that the propeller

slipstream covers a larger percentage of the wing area. This influences the wings’ aerodynamics

and potentially helps to delay stall. To investigate this behavior on the PingWing, the stall speed

was tested both with power on as well as with motor powered off. In the latter case, the MAV was

fitted with a folding propeller and the testing was conducted by flying at max speed, stopping the

motor and gliding at constant altitude until the nose dropped due to stall. The accelerometer data

helped identify the exact time of the stall in the data log. This was repeated several times in order

to acquire a mean value. The result is presented in Table 3.

Table 3. Propeller slipstream effect on CL max.

It should be mentioned that the power off stall speed is conservatively read from the data in order

to compensate for any possible lag in the pitot system. For the power on case, the CL was corrected

for the inclination of thrust vector using the crude angle of attack measure calculated from the

accelerometer data. Clearly, the propeller slipstream has a positive effect on maximum lift.

5. DISCUSSION

The results of the flight testing have provided interesting data. The data indicates that varying

turbulence intensities within the ABL have little to no effect on MAV performance. This was

unexpected. When observing the MAV flying in the wind, it was seen to be violently thrown

around and substantial control inputs were constantly required to maintain the flight path. The

instinctive feeling was that the abuse from the turbulence would notably reduce the MAV

performance. The results from Flight 2 show that for the range of turbulence in which it would be

realistic to operate, it is not possible to measure the difference in performance. For higher CL, it

even appears as if the performance is slightly increased in turbulence, although this is probably

within the error margin of the measurements. For the chaotic turbulence along the ridge line in

Flight 3, the performance did decrease but since the airspeed measurements cannot be guaranteed

and the turbulence level is not representative of any realistic scenario, that result is of less

importance.


14

Several interesting conclusions can be drawn from the results. Looking at the accelerometer

data in Figure 11, it is clear that the flow fluctuations occur at a rate where the aerodynamics are to

be classified as unsteady. One way to quantify the degree of unsteadiness in an airstream is the

reduced frequency parameter, k=ωc/2v, where ω is the angular frequency of the fluctuating

airstream, c is the chord of the wing and v is the free stream velocity. According to Leishman [18],

the unsteady aerodynamic effects can generally be ignored for reduced frequencies in the range of

0 < k < 0.05. Above this limit the flow is to be categorized as unsteady. Furthermore, at reduced

frequencies of 0.2 and above, the aerodynamics are to be considered highly unsteady and unsteady

effects will begin to dominate the air loads. If a spectral density analysis is made of the

accelerometer data of the most relevant case, Flight 2, it is seen that the most significant

fluctuations occur in the range of 5 Hz. This corresponds to k=0.21, i.e. highly unstable. With this

in mind, it is surprising that no difference in performance was noticed. On the other hand, the

classification using reduced frequency does not account for the amplitude of the fluctuations,

which reasonably should also play an important role. The vertical acceleration of Flight 2 is at

maximum fluctuation between 0 and 2 g. At a speed of 22m/s this corresponds to a CL variation of

0-0.22, or an angle of attack variation of 4.1 degrees, following the result in Figure 14. This

variation in angle of attack is not excessive and is probably not enough to alter the vehicle’s drag

characteristics significantly by itself.

An interesting comparison can be made by assuming steady aerodynamics and simulating the

potential drag increase in turbulence using the logged accelerometer data. This was done for Flight

2 and the accelerometer data in Figure 12. As a reference of lift to drag characteristics in steady

flow, the result from the flight in calm air was used. Averaged over all data points this resulted in

an increase in drag coefficient of 2.3%. This would hardly be noticeable in the data log. In this

calculation, however, no account has been taken of the drag induced by the relatively large side

forces caused by the turbulence. The induced drag due to side force is difficult to predict. A

conservative estimate can be made by assuming that all of the side force is generated by the

vertical stabilizers and then using standard equations for lift induced drag to compute the side

force induced drag. The vertical stabilizer is a clean aerodynamic surface and should provide a

much better “side force to drag ratio” then the actual vehicle has; these estimates should therefore

be seen as the minimum drag increase. Based on the Y-axis accelerometer data from Figure 12,

this approximation resulted in an average drag increase, due to fluctuating side force, of 12%. In

total, if steady aerodynamic theory were valid for turbulent conditions, the average drag

coefficient of Flight 2 should at a minimum be 14.3% higher than for Flight 1. This is for the flight

velocity of 22 m/s and would increase further for lower velocities. Clearly, this would have been

seen in the data and thus the unsteady aerodynamic effects do seem to be in effect. What is

interesting it that it appears as if the unsteady effects benefit the MAV to the point that the drag

generated by the constant disturbance in attitude and flight path is offset by an energy gain from

the turbulence. The reason for this result is not completely clear and leaves room for speculation.

The possibility to extract energy from an oscillating airflow was first demonstrated in

experimental measurements by Katzmayr [6]. Later analysis by Phillips [19] and Ribner [20]

theoretically supports Katzmayr’s results, but points out that for the turbulence levels found in the

atmosphere the potential energy gains are very limited. Phillips estimates a manned glider aircraft

to potentially see a thrust increase of approximately 10-20% of its drag, but also notes that in

reality no performance gains in flight through turbulence have been observed. For a MAV flying

in the ABL, the turbulence is comparatively stronger and, as pointed out by Langelaan and

Bramesfeld [22], it opens up the possibility for greater energy gains according to Phillips’ theory.

More recent work has explored this possibility for smaller UAVs, of conventional wing tail

configuration, either by using active control methods to optimally adjust the vehicles instantaneous

angle of attack for maximum energy gain [21-23], or by tuning aeroelasticity and structural

dynamics for the wing to naturally flex in order to optimally harvest the energy [24,25]. Another

area of energy harvesting in atmospheric winds, which has some similarities to the energy gain in

turbulence, is the phenomenon called dynamic soaring. Dynamic soaring can be described as

energy extraction by cyclic maneuvering between layers of different wind speed within the earth

boundary layer. It is an active research area that shows potential to substantial energy gains for

small UAVs [26-28]. In the case of the PingWing MAV, it appears as if it passively manages to


exploit energy from turbulence.

If passive energy gain from turbulence is possible purely from the MAV’s aerodynamic

shaping, it opens up a new perspective on MAV design and leads to many related research

questions. How should a MAV be optimized in order to best perform in the turbulent environment

of the real world? There may for instance be conflicting requirements between optimizing for low

Reynolds numbers and optimizing for energy gain in turbulence. At lower Reynolds numbers, thin

airfoils perform better, but on the other hand propulsive effects in turbulence are greater for

thicker airfoils [6]. Can traditional stationary theory be used for MAV design? The present results

are inconclusive. On the one hand, they show that stationary aerodynamic theory should not apply

for flight in the turbulent environment of the ABL while on the other hand, on the tested MAV, the

net effect of turbulence is negligible compared to the flight in smooth air. How general would this

result be on other MAV platforms? A further interesting study would be to compare wind tunnel

measurements with flight test data, as even in the calmest of conditions there is still some

fluctuating wind within the ABL. Would the degree of static stability influence the performance in

turbulence? A neutrally stable aircraft is in general considered the best for a sensor platform in the

sense that its attitude (pitch, roll yaw) will be minimally impacted by turbulence. On the other

hand, a MAV of flying wing configuration has very low pitch inertia and if longitudinally stable it

will to some extent self-orient itself with the direction of the flow, which probably influences the

possibility to gain energy from turbulence.

In the presented results no error margins have been included. The measurement error can be

divided into two types: the sampling error and bias error. The sampling error was calculated

assuming a normal distribution of the data points. In order to compensate for the noisy

measurements in the turbulent flight, a significant number of data points were used to form the

mean value and due to this the error analysis gave 95% confidence bounds that were so small that

it did not make any sense to include them in the plots of the results. The bias error could have been

estimated but since the data is of a comparative nature it would not contribute to the end result.

Looking at the spread of the data, there are however some obvious uncertainties as regards the

mean values. These errors are likely due to the stochastic nature of the ABL, or piloting precision,

and these errors are very hard to estimate. The instantaneous performance could be seen to both

improve and decrease due to the turbulence and this variation probably did not follow a normal

distribution within the time frame of the measurements. The precision in the polynomial curve fits

is also dependent on the number of points included, as well as their spread. For this reason, a more

systematic method of setting the throttle levels, other than letting the pilot arbitrarily do the

control, would have been better.

The described method of testing propellers is only peripheral to this work, but it has been very

successful and deserves some discussion. Overall, the method has given very good results. It is an

excellent wind tunnel substitution at a fraction of the cost. A drawback of the method is obviously

that testing can only be carried out when the weather allows it. A benefit is that the environment is

more representative then the one found in a wind tunnel and no correction for wall interactions etc.

is needed. Experience shows that care should be taken to carry out the testing on a road aligned

with the wind direction even if conditions are close to calm. The described test rig has worked

well but could be further improved. The chosen data logger suffers from low sampling rate. A

higher logging frequency would shorten the time needed for each test run. To minimize the

problem of flow inclination not always being perpendicular to the propeller disk, the entire

measuring rig could be made to swivel on the tripod similar to a weather vane. On the other hand,

the non-zero flow inclination experienced with the rigid solution can also be used to advantage. In

combination with the automated data reduction technique, a large amount of testing could fairly

easily be done to build a database of propeller characteristics, where flow inclination is included as

a parameter in addition to rpm and advance ratio.

6. SUMMARY AND CONCLUSIONS

A MAV was flight tested in varying wind conditions in order to investigate the effect of

turbulence on performance. As a first step, the thrust and torque characteristics of the MAV’s

propeller were determined in a test rig carried on top of a moving car. The propeller data was then

used to determine the drag characteristics of the MAV flown in calm as well as highly turbulent


16

wind. Surprisingly, the lift to drag ratio averaged over time exhibited little noticeable variation for

the different turbulence levels. The MAV was highly impacted by the turbulent wind, which

resulted in a constantly disturbed flight path as well as significant oscillations in yaw. If stationary

aerodynamic theory is assumed this should have resulted in a noticeable increase in drag, yet

performance was not affected. This indicates that the MAV to some extent benefits from the

turbulence to the point that the increased drag due to interference is offset by an energy gain from

the turbulence. Moreover, accelerometer data of the flight confirms that the flow around a MAV

flying in the lower region of the atmospheric boundary layer is dominated by unsteady

aerodynamic effects.

REFERENCES [1] Kaimal, J.C. and Finnigan, J.J., Atmospheric Boundary Layer Flows: Their Structure and Measurement, Oxford

University Press, 1994.

[2] Stull, R.B., An introduction to boundary layer meteorology, Kluwer Academic Publishers, 1988.

[3] Liu, H., Wind Engineering: A Handbook for Structural Engineers, Prentice Hall, 1991.

[4] Watkins, S., Thompson, M., Loxton, B. and Abdulrahim, M., On Low Altitude Flight Through The Atmospheric

Boundary Layer, International Journal of Micro Air Vehicles, Volume 2, Number 2, June 2010, pp 55-68.

[5] Sytsma, M.J. and Ukeiley, L., Low Order Turbulence Modeling Methods for MAVs Flight Environment, AIAA Atmospheric Flight Mechanics Conference, Toronto, Ontario Canada, Aug 2010.

[6] Katzmayr, R., Effect of Periodic Changes of Angle of Attack on Behavior of Airfoils, NACA-TN-2371, 1922.

[7] Watkins, S., Loxton, B.J., Milbank, J., Melbourne, W.H. and Abdulrahim, M., Wind-Tunnel Replication of Atmospheric Turbulence with an Emphasis on MAVs, 46th AIAA Aerospace Sciences Meeting and Exhibit, Reno,

NV, USA, Jan 2008.

[8] Amadori, K., Lundström, D. and Krus, P., Automated Aesign and Fabrication of Micro-Air Vehicles, Accepted for publication in Journal of Aerospace Engineering, Proceedings of the Institution of Mechanical Engineers Part G

[PIG], DOI: 10.1177/0954410011419612, 2011.

[9] Lundström, D., Amadori, K. and Krus P., Automation of Design and Prototyping of Micro Aerial Vehicles, 47th AIAA Aerospace Sciences Meeting and Exhibit, Orlando, FL, USA, Jan. 2009.

[10] Ostler, J.N., Bowman, W.J., Snyder, D.O. and McLain, T.W., Performance Flight Testing of Small, Electric

Powered Unmanned Aerial Vehicles, International Journal of Micro Air Vehicles, Volume 1, Number 3 , November 2009, pp 155-171.

[11] UIUC propeller database: www.ae.illinois.edu/m-selig/props/propDB.html.

[12] Lundström, D., Amadori, K. and Krus P., Validation of Models for Small Scale Electric Propulsion Systems, 48th AIAA Aerospace Sciences Meeting and Exhibit, Orlando, FL, USA, Jan. 2010.

[13] Gracey, W., Measurement of Aircraft Speed and Altitude, NASA Reference Publication 1046, Langly Research

Center , Hampton, Virginia, USA.

[14] DeYoung, J., Force and Moment Derivatives due to Propellers of Arbitrary Configuration Inclined With Respect To

Free Stream, AIAA General Aviation Aircraft Design & Operations Meeting, Wichita, Kansas, USA, May 1964.

[15] Lesley, E.P., Worley, G.F. and Moy S., Air Propellers in Yaw. NACA report No. 597.

[16] Conte, G., Hempel, M, Rudol, P., Lundström, D., Duranti, S., Wzorek, M., Doherty, P., High Accuracy Ground

Target Geo-Location Using Autonomous Micro Aerial Vehicle Platforms, AIAA Guidance, Navigation and Control

Conference and Exhibit, Honolulu, HI, USA, Aug 2008.

[17] Wieringa, J., Davenport, A.G., Grimmond, C.S.B. and Oke, T.R., New revision of Davenport roughness

classification, Proceedings of the 3rd European & African Conference on Wind Engineering, Eindhoven,

Netherlands, July 2001.

[18] Leishman, J.G., Principles of Helicopter Aerodynamics, 2nd edn., Cambridge University Press, 2002.

[19] Phillips, W.H., Propulsive Effects due to Flight through Turbulence, Journal of Aircraft. Vol. 12, No.7, 1975.

[20] Ribner H.S., Thrust Imparted to an Airfoil by Passage Through a Sinusoidal Upwash Field, AIAA Journal, Vol. 31, No. 10, Oct 1993.

[21] Depenbusch, N.T. and Langelaan, J.W., Receding Horizon Control for Atmospheric Energy Harvesting by Small UAVs, AIAA Guidance, Navigation and Controls Conference, Toronto, Canada, August 2010.

[22] Langelaan J.W. and Goetz Bramesfeld G., Gust Energy Extraction for Mini- and Micro- Uninhabited Aerial

Vehicles, 46th AIAA Aerospace Sciences Meeting, Reno, Nevada, USA, 2008.

[23] Patel, C.K. and Kroo, I.M., Theoretical and Experimental Investigation of Energy Extraction from Atmospheric

Turbulence, 26th International Congress of the Aeronautical Sciences, Anchorage, Alaska, USA, Sep 2008.


[24] Ironside, D.J., Bramesfeld, G. and Schwochow, J., Modeling of Wing Drag Reductions Due to Structural Dynamics in Atmospheric Gusts, 28th AIAA Applied Aerodynamics Conference. Chicago, Illinois, USA, 2010.

[25] Bramesfeld, G., Ironside, D. J., and Schwochow, J., Simplified Modeling of Wing-Drag Reduction due to Structural

Dynamics and Atmospheric Gusts, AIAA 26th Applied Aerodynamics Conference, Honolulu, HI, USA, Aug 2008.

[26] Boslough, M.B.E., Autonomous Dynamic Soaring Platform for Distributed Mobile Sensor Arrays, Tech. Rep.

SAND2002-1896, Sandia National Laboratories, California, USA, June 2002.

[27] Lawrance, N.R.J. and Sukkarieh, S., A guidance and control strategy for dynamic soaring with a gliding UAV, 2009 IEEE International Conference on Robotics and Automation, Kobe, Japan, May 2009.

[28] Mears, M., Energy Harvesting for Unmanned Air Vehicle Systems using Dynamic Soaring, 50th AIAA Aerospace

Sciences Meeting, Nashville, TN, USA, Jan 2012.

Documents

Testing of Atmospheric Turbulence Effects on the ...€¦ · Drag coefficient Zero lift drag ... Over flat uniform terrain it typically extends to a height of 500-1000 ... and yaw